1P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Applications of PBPK Modeling in Preclinical Drug Development

Micaela ReddyDMPK Department, Modeling & Simulation Group

Roche Palo AltoMarch 30, 2009

Presented at the Hamner Institutes for Health Sciences Course:Physiologically Based Pharmacokinetic (PBPK) Modeling in Drug Development and EvaluationApril 610, 2009

2P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Goals

Present examples of multiple ways that PBPK modeling can be used to address key issues for preclinical projects

Describe how PBPK models can be used to generate hypotheses and prioritize additional experiments

Provide a method for developing physiologically based oral absorption models and discuss the factors that might impact oral absorption for BCS Class 1, 2, 3 and 4 compounds

Illustrate how PBPK/PD modeling can be used to assess the likelihood that a compound will be efficacious in the clinic and to characterize the therapeutic window

3P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

PBPK models have many applications in preclinical drug development because of the need to integrate data In silico estimates

pH Dependence of Solubility

0.001

0.01

0.1

1

10

0 1 2 3 4 5 6 7 8 9 10

pH

l

o

g

(

S

o

l

u

b

i

l

i

t

y

)

[

m

g

/

m

L

]

CalculatedMeasuredFitted

Solubility

Pro

tein

bin

din

g

Donor Acceptor

time

conc.

Intestinal

permeability P-gp In vivo

PK

Normal Medium with Calcium

Cell Lysis andSamplingUptakeUptake

Ca2+-free Medium with 1 mM EGTA

Active processesM

etab

olism

Conc

h

formulation

T. Lav

4P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Outline

PBPK models for generating hypotheses Oral absorption modeling PBPK modeling for candidate selection Predicting exposure window using PBPK/PD

5P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

PBPK modeling is an evolving process

New paradigm: See if you can predict in preclinical species. If you cannot, perform appropriate experiments to fill the knowledge gap.

Mechanistic knowledge and availability of key data determine the confidence you can have in model predictions.

6P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Fundamental value of modeling

Models allow you to integrate multiple sources of data and information with a mechanistic description based on your fundamental understanding of the system.

If the model prediction is correct, it is useful. Instead of making predictions based on rat data, you can

take it a step further. After testing your understanding of the system with rat, you can extend that knowledge to make predictions in man.

If the model prediction is incorrect, it is also useful. The disconnect between the model and data suggests

your understanding of the system is flawed. The model can be used to rule out hypotheses that are

inconsistent with the data and determine additional experimentation that is most likely to be informative.

7P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Example: Lack of IVIVC from nonlinear CL

Experiments:

Rat and dog bile-duct cannulated studies to better understand clearance

mechanisms: No significant biliary secretion

Using lower concentration of drug (0.1 M instead of 1 M) for clearance

measurement in vitro to prevent saturation of metabolic enzymes in

hepatocyte or microsome to obtain accurate intrinsic clearance

Problem: Modeling indicated lack of good

IVIVC

Rat PO Profile at 3 mpk

0.1

1

10

100

1000

0 10 20 30

Time Post Dose (h)C

o

n

c

e

n

t

r

a

t

i

o

n

(

n

g

/

m

L

)

PBPK SimulationMeasured PK

D. Lu

8P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Outcome: IVIVC achieved for the compound within 2-fold uncertainty using newly obtained microsome clearance. Proceeded with PBPK simulation in human.

Rat PO Profile at 3 mpk

0.1

1

10

100

1000

0 5 10 15 20 25 30

Time Post Dose (h)

C

o

n

c

e

n

t

r

a

t

i

o

n

(

n

g

/

m

L

)

PBPK SimulationMeasured PK

D. Lu

9P

h

a

r

m

a

c

e

u

t

i

c

a

l

sExample: Impact of enterohepatic circulation (EHC) on PK

Courtesy of Dr. Werner Rubas

10

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sFor this compound, CL is overpredicted and t1/2 is underpredicted by the in vitro model parameters

Hypothesis: Enterohepatic circulation of parentEstimation of fraction EHC = (CLobs CLpred)/CLobs = 0.25

0.1

1

10

100

1000

0 5 10 15 20 25 30

Time [h]

C

o

n

c

.

[

n

g

/

m

L

]

ObservedModeled

W. Rubas

11

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sInclusion of predicted EHC (25%) improves simulation

Experiment using bile duct cannulated rats: Fraction EHC was determined to be 0.33.

0.1

1

10

100

1000

0 5 10 15 20 25 30Time [h]

C

o

n

c

.

[

n

g

/

m

L

]

Observed

Modeled w/o EHC

Modeled w EHC

W. Rubas

12

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

PK profiles are well simulated when enterohepatic circulation is included (33%)

0.1

1

10

100

1000

0 5 10 15 20 25 30Time [h]

C

o

n

c

.

[

n

g

/

m

L

]

Observed

Modeled

Predicted AUC and t1/2 are within 10 % of observed

W. Rubas

IV 0.5 mg/kg PO 2 mg/kg

0.1

1

10

100

1000

0 5 10 15 20 25 30Time [h]

C

o

n

c

.

[

n

g

/

m

L

]

ObservedModeled w/o EHCModeled w EHC

Cmin prediction is improved

13

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Outline

PBPK models for generating hypotheses Oral absorption modelingValidating the modelUsing the model

Compound prioritization based on early preclinical data Predicting exposure window using PBPK/PD

14

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sRecently, several powerful PBPK approaches have been developed to predict oral absorption.

Example: ACAT model in GastroPlus

The model can also include degradation of compound passing through the GI tract, the

impact of transporters, and differences in permeability with region.

Species differences in regional pH, compartment volume, etc. are included in the model.

Other examples:

ADAM (advanced dissolution, absorption & metabolism) model (Jamei et al., in press),

GI tract described as tube with spatially varying properties (Willman et al., 2003)

15

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Clinical lead

How can M&S increase efficacy and speed in drug discovery?

Front load

Critical parametersWhich one to focus on first?

Data (in vitro/in vivo)

Metabolic stability?

Protein binding?

Solubility?

Validation model (few compounds)

Not critical

Sensitivity analysis

Stop experiment

Multiple series

Screening funnel

16

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

PBPK modeling for oral absorption

PBPK modeling approaches for predicting human oral absorption can be used to simulate absorption as a function of time the amount absorbed from various regions of the GI tract plasma concentrations as a function of time

Approach: Develop a model in a preclinical species and check it against

whatever data is available (e.g., po SDPK data) Determine how to improve the model based on mismatches

between model and data; perform critical experiments to elucidate key mechanisms

Develop the human model based on what was learned in the preclinical species

17

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Applications of oral absorption modeling

Predicting bioavailability Evaluate potential absorption and bioavailability for large

number of compounds with very limited data Determine the sensitivity of bioavailability and PK to various

input parameters Maximum absorbable dose Predicting whether a food effect is expected Determining an appropriate formulation strategy

18

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sAn absorption model can be developed using permeability and solubility data

Based on solubility data for range of pHs

In silicoSolubility factor

pH-solubility profile, including Fessif/Fassif

Aqueous; HT Fessif/Fassiffor low solubility compounds (e.g., < 20 ug/ml aqu. sol.)

Solubility @ reference pH

Cell-based permeability assay (e.g., MDCK, Caco-2) or PAMPA

In silicoOptional: cell-based permeability assay (e.g., MDCK, Caco-2) or PAMPA

Permeability (human Peff)

MeasuredIn silicopKas

Measured Log D-pH profileIn silicocLogP / Log D @ reference pH

Late in development (CLS, EIGLP) Early in development (LO) Parameter

Physicochemical properties

19

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Outline

PBPK models for generating hypotheses Oral absorption modelingValidating the modelUsing the model

PBPK modeling for candidate selection Predicting exposure window using PBPK/PD

20

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

To check your absorption model

Simulation results can be compared to: Bioavailability for a single experiment Bioavailability as a function of dose Fraction absorbed into the portal vein Plasma time-course concentrations (requires a PK study)

Before using absorption modeling to make decisions, it is good practice to check the model results against PK data at multiple doses and in one or more species if possible.

21

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Using absorption modeling to predict bioavailability

F = Fabs x Fh x Fg x where Fabs is the fraction that is absorbed into enterocytes Fh is the fraction that makes it through the liver Fg is the fraction that makes it through the enterocytes

without being metabolized There could be additional sources of loss (e.g., metabolism by

microflora in the gut, instability in the stomach, etc.). Absorption modeling provides a prediction of fraction

absorbed, but to predict in vivo bioavailability, first-pass processes must also be incorporated.

22

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Definition of Bioavailability (F) Bioavailability (F) is the fraction (or percent) of administered dose

that reaches systemic circulation intact. Bioavailability is typically calculated

by comparing AUC from iv PK data to AUC from po PK data.

Often bioavailability is < 1 because loss can occur from factors like incomplete absorption (e.g., from low permeability or solubility) first pass extraction by the gut and liver chemical degradation in the lumen

Sometimes estimated F is > 1 than because of factors such as saturable metabolism enterohepatic circulation incomplete administration of iv dose

iv

iv

po

po

DAUC

DAUC

F =

23

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Comparing Fabs to Bioavailability

One mechanism that often has a large impact on bioavailability is first-pass hepatic extraction.

At a minimum, the effect of first-pass extraction (Eh) should be combined with Fabs for estimating bioavailability. (Assume Fg=1 if no phenotype data are available.)

Fh = fraction that makes it past the liver, 1 - EhFabs = fraction absorbed

When can Fabs be compared directly to bioavailability? For compounds that are cleared by mechanisms other than

hepatic metabolism or biliary excretion For low-clearance compounds (CL < 0.1 x QL)

F = Fh x Fabs

24

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Methods for calculating Fh

Fh can be calculated as: Fh = 1 Eh = 1 CL / QL Eh can be calculated from hepatocyte or microsome in vitro data Eh can also be estimated using CL from an iv SDPK study (CL / QL)

Caution: Is the mechanism of CL hepatic metabolism? Which method to use? Do scaled hepatocyte or microsome data:

predict IV CL? Either method can be used. underpredict IV CL? Metabolism is probably not the only

mechanism of CL, and so scaled hepatocyte or microsome data should be used.

overpredict IV CL? Most CL is probably from metabolism, and the IV CL can be used as the more accurate CL estimate.

CL = CLplasma / BPR where BPR = blood to plasma ratioIn some cases (e.g., high extraction compounds that do not partition into red blood cells) it is critical to have BPR data for interpretation of PO PK data!

25

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Checking model prediction of bioavailability

0.0290.370.0770.388.41.526.1 (Aq)Compound 40.130.370.350.388.41.5232.9 (Fa)Compound 40.230.370.620.380.8713.2 (w/ elac.b)32.9 (Fa)Compound 4

0.240.310.780.460.7715.338.7 (Fa)Compound 30.0240.310.0790.460.7715.32 (Aq)Compound 30.190.340.550.341.116.321.2 (Aq)Compound 20.220.230.960.261.213.2145 (Aq)Compound 1

FabsxFhFhFabsFERCaco2 Peff, cm/sx10-6

Solubility, g/ml a

a Aq = aqueous solubility at pH=6.5. Fa = FaSSIF solubility at pH=6.5.

Selected drug properties and F from rat SDPK data with Fabs fromGastroPlus and Fh from an appropriate method.

b Elacridar is a P-gp inhibitor.

26

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Checking bioavailability prediction for a range of doses

Multiple mechanisms affect exposure as a function of dose.However, in terms of Fabs, one would generally expect thefollowing: BCS Class 2 and 4 may exhibit dose-proportional or less-

than-dose-proportional increases in AUC and Cmax as the dose increases. If the model predicts that dose-limited absorption occurs

(i.e., simulated Fabs decreases with increasing dose), there should be a less-than-dose-proportional increases in AUC and Cmax as the dose increases in po SDPK studies.

BCS Class 1 and 3 may exhibit dose-proportional or greater-than-dose-proportional increases in AUC and Cmax as the dose increases in po SDPK studies.

27

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Checking fraction absorbed into the portal vein *

The Fabs into the portal vein will not reflect the effects of first-pass hepatic metabolism (although other first-pass processes, such as metabolism by enterocytes, might still have an impact).

Therefore, Fabs into the portal vein can be used more easily to check the model prediction of Fabs.

700 mg ITMN-191 PO Dose

0

50

100

150

200

250

300

0 4 8 12 16 20 24 28

time, h

C

u

m

u

l

a

t

i

v

e

A

a

b

s

,

m

g

KFAKFBKFCKFDGP sim

100 mg/kg PO dose

* Useful for compounds with high hepatic extraction!

28

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Calculation: Amount Absorbed The amount absorbed as a function of time can be calculated from

the concentrations in the portal and peripheral vein.

d Aabs = Qpv x (Cpv Cp) x BPRdt

Aabs = amount absorbedt = timeQpv = portal vein blood flow rate (~85% of liver blood flow)Cpv = plasma concentration in portal veinCp = plasma concentration in peripheral veinBPR = blood-to-plasma ratio

29

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Checking the model against plasma time-course concentrations

Rat, 2 mg/kg in Suspension

1

10

100

1000

10000

0 4 8 12 16 20 24time, h

C

p

,

n

g

/

m

l Rat 1Rat 2Rat 3Sim

Rat, 20 mg/kg in Suspension

1

10

100

1000

10000

100000

0 4 8 12 16 20 24

time, h

C

p

,

n

g

/

m

l Rat 1Rat 2Rat 3Sim

Monkey, 2 mg/kg

100

1000

10000

100000

0 4 8 12 16 20 24

time, h

C

p

,

n

g

/

m

l KF 1KF 2KF 3Sim

Monkey, 20 mg/kg

100

1000

10000

100000

0 4 8 12 16 20 24

time, h

C

p

,

n

g

/

m

l KF 1KF 2KF 3Sim

30

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sApproach: Reduce confounding factors by using a compartment model for systemic PK.

For rat PO PK, first use a compartment model to simulate iv PK to remove one source of uncertainty and check the absorption prediction. Assumption: linear kinetics First-pass Eh will need to be calculated and included After checking absorption model, redo calculation to check

the PO PK profile using the PBPK model.2 mg/kg po PKChange to po

exposure, parameterize

oral absorption model

0.5 mg/kg iv PK

Compartment model for systemic PK

31

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sNext, switch to the PBPK model for systemic PK to verify that the model is still predictive.

After checking the absorption model, redo the calculation to check the PO PK profile using the PBPK model.

2 mg/kg po PK0.5 mg/kg iv PK Change to po exposure,

parameterize oral absorption

modelPBPK model for systemic PK

32

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sWhat do you do if the model does not match the data?

Several avenues of exploration User error

Double check input, paying close attention to units Are solubility, permeability, and compound properties

entered correctly? Inappropriate input for BCS Class 2, 3 or 4

BCS Class 2, 4: need appropriate solubility data BCS Class 3, 4: carefully consider best way to estimate

Peff Important mechanism impacting absorption or PK not yet

identified (e.g., metabolism by gut, nonlinear metabolism) Need to design appropriate experiments

33

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sAbsorption models for BCS Class 1 compounds tend to be reliable, but issues can still arise.Example: Dissolution data was necessary to simulate absorption of diltiazem (BCS Class I) The initial model overpredicted Cmax and underpredicted tmax.

Dissolution data was required to simulate absorption. Input: Dissolution data, solubility = 3000 g/ml @pH = 6.4, PAMPA

Peff = 1.9x10-6 cm/s, hepatic extraction = 54%

Dog PO PK, Teva IR 120 mg Tablet

0

400

800

1200

1600

0 4 8 12 16 20 24

Time (hours)

C

p

,

n

g

/

m

L

Dog 1Dog 2Dog 3MeanSim, genericSim, disso data

0

20

40

60

80

100

0 50 100 150 200time (min)

Observed

Predicted

Formulation=IR, Teva

Weibull function

34

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

For human simulation, dissolution data and first-pass extraction by the gut had to be incorporated.

Cardizem IR - 120 mg dose

0

20

40

60

80

100

120

0 4 8 12 16 20 24

time, h

C

p

,

n

g

/

m

l No GFJGFJSim, no GFJSim, GFJ

Data from Christensen et al. 2002

0

20

40

60

80

100

0 500 1000 1500time (min)

Observed

Predicted

Formulation=IR, Cardizem

Weibull function

* Most PBPK software packages include CYP expression in the gut and can predict Fg if reaction phenotyping data is available.

One way to determine the impact of first pass extraction by the gut is to do a grapefruit juice (GFJ) study because it inhibits CYP3A4 in the gut (Yang et al. 2007). Fg < 1 for many CYP3A4 substrates, e.g., midazolam (Fg = 0.57), felodipine (Fg = 0.58), and saquinavir (Fg = 0.67).

35

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Considerations for dissolution data

Dissolution data might improve your prediction, depending on how the study is designed and whether dissolution is rate limiting.

Considerations for study design: Sink conditions Physiologically relevant

media (FeSSIF, FaSSIF, FeSGF, FaSGF)

Two-stage design (e.g., pH 1 and then pH 6)

BCS Class II compound

DissolvedAbsorbed

Key plot for diagnosing whether dissolution might be rate-limiting

36

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sBCS Class 2Low solubility, high permeability

For BCS Class 2 solubility is a very sensitive model parameter, and appropriate (i.e., physiologically relevant) solubility data is critical.

FeSSIF and FaSSIF solubility data have been shown to dramatically improve the predictive accuracy of the absorption model for low-solubility compounds.

If the dosing solution contains solubility enhancers, solubility data with solubility enhancers might also be appropriate.

Different batches of drug can have very different solubilities. A mismatch between Fabs in a po SDPK study and the model prediction of Fabs could be because the compound administered to the animals and the compound used to determine the solubilitywere from different batches.

Efflux transporters might be an issue.

37

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sVerify that sufficient data are available to set solubility

If a compound has a physiologically relevant pKa, carefully consider if solubility data have been generated for a sufficient range of pH values.

38

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sEarly stage prediction of bioavailability in rats for BCS Class 2 compounds

Used to verify understanding of factors impacting series %F

Courtesy of Q. Hu

y = 1.0x

0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

40.0

45.0

0 5 10 15 20 25 30 35 40 45

%F (measured)

%

F

(

c

a

l

)

Compounds:-CLogP, pKa from ADMET Predictor;-Dose form: IR suspension;-Solubility: phosphate (if >20 ug/mL) and FaSSIF -Particle density 1.2 g/mL-particle size 25 m-Peff from Caco2 AB

Physiology: Rat (fasted), Opt LogD ModelSimulation: 24hFirst-pass extraction: In vivo IV clearance

12 within 2-fold2 over-predicted2 under-predicted

39

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sBCS Class 3Low permeability, high solubility

For BCS Class 3, the effect of transporters may have a large, and not easily predicted, impact on PK. At low doses, the impact of transporters will be largest. At high doses, transporters could saturate. Solution: include saturable transport into the model.

For estimating Peff for Pgp substrates, Caco2 (A>B) data with elacridar is likely more predictive. But for lower solubility or permeability compounds, Caco2 (A>B) data without elacridar couldbe more predictive. Looking at the Fabs calculation using both Peff values could provide a reasonable range of values.

In the intestines (duodenum, proximal jejunum scrapings), CYP3A4was the most abundant P450 (80% of intestinal P450), followed byCYP2C9 (15%) (Paine et al. 2006). Gut metabolism may be most important for low permeability CYP3A4 substrates (Yanget al. 2007).

40

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

The gut expresses many efflux and uptake transporters. Expression levels can vary depending on location in GI tract.

Figure from: Predicting drug disposition, absorption/elimination/ transporter interplay and the role of food on drug absorption. Custodio JM, Wu CY, Benet LZ(2008). Adv Drug Deliv Rev 60(6): 717-733.

41

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Early stage prediction of bioavailability in rats for BCS Class 3 compounds

Used to verify understanding of factors impacting series %F

Data: In silico cLogP, pKa; Caco2 Peff with no inhibitor, THESA solubility, Eh calculated using Mx data

Courtesy of S. Larrabee

5 over-predicted8 within 2-fold

42

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sBCS Class 4Low solubility, low permeability

For BCS Class 4, the issues that occur with BCS Class 2 and 3 might both occur.

For BCS Class 4, it might be important to do more checking than for the other BCS classes, and to be more careful with interpretation of results, particularly if transporters are involved.

Absorption modeling can still be useful, but results should be carefully interpreted. The key is to interpret more qualitatively than quantitatively unless you have validated the model appropriately.

43

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Outline

PBPK models for generating hypotheses Oral absorption modelingValidating the modelUsing the model

PBPK modeling for candidate selection Predicting exposure window using PBPK/PD

44

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Examples from the literature

Peters (2008): Assess any lineshape mismatch between simulated and observed oral profiles to gain mechanistic insights into processes impacting absorption and PK for example, drug-induced delays in gastric emptying and

regional variation in gut absorption. Jones et al. (2006): GastroPlus human oral absorption

models were established for six compounds. Food effects were predicted for a range of doses and compared to the results from human food effect studies. In general, the models were able to predict whether a food

effect would be major (i.e., for two compounds) or minor (i.e., for four compounds).

45

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Example: Is a CR formulation feasible?

CR formulation development is expensive, but can be necessary for a compound with a short t1/2 in humans.

Extensive PK experimentation is typically performed to determine whether a CR formulation is feasible (i.e., whether the compound can be absorbed in the colon) before resources are spent on the effort. Surgically modified animals with cannulas for

administering compound to various parts of the GI tract Cross-over IV and PO PK studies for deconvolution to

determine the amount absorbed as a function of time EnterionTM capsule, Pharmaceutical Profiles

Modeling can help to focus and reduce the experimentation.

46

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sSite-of-absorption studyExample: Furosemide (BCS Class 4)

PO

IJ IC

Based on the po PK profiles alone, it is not clear where absorption occurs.

Simulations performed with GastroPlus demo database.

Simulated Human Site-of-Absorption Data100

mg Furosemide

0

0.2

0.4

0.6

0.8

1

0 4 8 12 16 20 24

time, h

C

p

,

u

g

/

m

l

POIJIC

47

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Deconvolution of po PK data (an empirical approach) can also be useful for understanding absorption.

If iv and po PK data are available, numerical deconvolution can be used to estimate the fraction of drug absorbed as a function of time. The results are sometimes noisy and difficult to interpret. Better results are obtained with a cross-over design so that

iv and po data are available in the same animal. This analysis will not necessarily result in clear information on

where drug is absorbed (i.e., results may be difficult to interpret).

For compounds that are well absorbed in the small intestines, the po data will have no information on the rate of absorption from the colon.

48

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sExample: Ketoprofen (BCS Class 1)

Simulation performed with GastroPlus demo database.

Simulation for humans administered a 50 mg dose of ketoprofen:

Most ketoprofen absorption occurs in the duodenum and jejunum. The po PK data contains no information on rate of absorption from the large intestines.

Simulations can provide useful information about whether deconvolution of po PK data will yield useful information.

49

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sExample: Should a CR formulation be developed for a compound with a short t1/2?

0

20

40

60

80

100

0 4 8 12 16 20 24

time, hC

u

m

u

l

a

t

i

v

e

A

a

b

s

,

m

g

0.31.21.73.4

0

20

40

60

80

100

0 4 8 12 16 20 24

time, h

C

u

m

u

l

a

t

i

v

e

A

a

b

s

,

m

g

0.31.21.73.4

For this compound, Caco2 Peff data had been measured and exhibited 10-fold variability.Simulations indicate that the intracolonic route in humans may result in 31-46% of the PO AUC. Sensitivity analysis shows that Peff is critical for making the prediction.

Peff (x10-6 cm/s) =

Peff (x10-6 cm/s) =PO IC

50

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sModel refinement with new data suggests that CR development would be very challenging.

Peff was a sensitive model parameter. Additional experimentation revealed that the compound had

permeability in the low end of the range. Refined simulations indicated that the IC route in humans

would likely result in 31% of the PO AUC or less. In a recent PharmaProfiles poster by Connor et al. (2008),

CR development is classified as very challenging for compounds with a relative bioavailability of < 30% upon administration to the colon.

Based on this assessment, CR development would likely be very challenging.

51

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Outline

PBPK models for generating hypotheses Oral absorption modeling PBPK modeling for a parent compound and metabolite PBPK modeling for candidate selection Predicting exposure window using PBPK/PD

52

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

PBPK For Candidate Selection

Aim: to use physiologically based simulation tools to combine theavailable PK and PD data and assist in selection of the optimal clinical candidate

Steps taken: Verification: compare simulated and observed PK in rat Predict PK: estimate human PK Predict PD: effective concentrations in human Estimation: clinical doses and exposures

Outcome: Balanced comparison of PK/PD incorporating variability

and uncertainty to be weighed with other factors (early tox, synthetic tractability, formulation ease etc)

N. Parrott

53

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

PBPK For Candidate Selection

0.000.100.200.300.40

fu% rat fu% man

0

2

4

6

ID90 rat(mg/kg)

010203040

Ki rat (nM) Ki human (nM)

00.5

11.5

Clint human hepatocytes [uL/min/M cells]0

10203040

CL Rat (ml/min/kg)

Available data for 5 compoundsPhysicochemicalIn vitro hepatocytes rat & manProtein binding rat & manIn vivo PK i.v. and p.o. in ratEffect versus concentration in rat

N. Parrott

54

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Simulated Cmax in human for a dose of 25mg including variability in both CL and V based upon in vitro and in vivo data

Mean with 95% limits shown

PBPK For Candidate Selection

050

100150200250300350400

R

O

C

H

E

_

1

R

O

C

H

E

_

2

R

O

C

H

E

_

3

R

O

C

H

E

_

4

R

O

C

H

E

_

5

C

m

a

x

[

n

g

/

m

L

]

N. Parrott

55

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Variability is based on the observed variability in the rat

1

10

100

1000R

O

C

H

E

_

1

R

O

C

H

E

_

2

R

O

C

H

E

_

3

R

O

C

H

E

_

4

R

O

C

H

E

_

5

C

o

n

c

f

o

r

9

0

%

r

e

v

e

r

s

a

l

[

n

g

/

m

L

]

PBPK For Candidate SelectionEffective concentration in human estimated from rat PD data rat and in vitro data in both species (Fu% rat & human, Ki rat & human)

N. Parrott

56

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

0.000.100.200.300.40

fu% rat fu% man

0

2

4

6

ID90 rat(mg/kg)

010203040

Ki rat (nM) Ki human (nM)

00.5

11.5

Clint human hepatocytes [uL/min/M cells]

010203040

CL Rat (ml/min/kg)

1

10

100

1000

Conc. for 90% effect inhuman (ng/mL)

020406080

Daily dose [mg]

Prediction in HumanSimulation

Multiple Discovery Data

Integratesmultiple PK and PD data

Aids rational and balanceddecision

Focuses on expected humanPK/PD

N. Parrott

57

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Outline

PBPK models for generating hypotheses Oral absorption modeling PBPK modeling for a parent compound and metabolite PBPK modeling for candidate selection Predicting exposure window using PBPK/PD

58

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sHypothetical examplePredicting therapeutic window for drug

with safety issue and nonlinear PK

A BCS Class II compound has dose-limited absorption. Efficacy for the compound is related to percent inhibition,

which can be described using a simple Emax model with EC50 = 40 ng/ml.

The compound has an off-target effect, and the PD for this effect also can be described using a simple Emax model with EC50 = 2000 ng/ml.

A simple modeling approach can be used to understand whether efficacy can be achieved while avoiding the PD that is a safety concern.

59

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

0

0.04

0.08

0.12

0.16

0 4 8 12 16 20 24time, h

C

p

,

u

g

/

m

l 20 mg40 mg60 mg80 mg100 mg

Even at low doses, dose-limited absorption is expected to occur.

Fabs = 31%

Fabs = 62%

60

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sDose-limited absorption might make it difficult to achieve efficacious Cp

For efficacy, a minimum of 50% inhibition is required.

020406080

100

0 4 8 12 16 20 24

time, h

E

f

f

i

c

a

c

y

,

%

o

f

E

m

a

x

20 mg40 mg60 mg80 mg100 mg

61

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sAt a 100 mg dose, the safety PD would likely be detectable.

The safety effect cannot be determined as different from the baseline for < 5% response, and so doses up to 60 mg might be acceptable.

01234567

0 4 8 12 16 20 24

time, h

S

a

f

e

t

y

P

D

,

%

c

h

a

n

g

e

f

r

o

m

b

a

s

e

l

i

n

e 20 mg40 mg60 mg80 mg100 mg

62

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

00.020.040.060.08

0.10.120.140.16

0 4 8 12 16 20 24

time, h

C

p

,

u

g

/

m

l

0

20

40

60

80

100

E

f

f

i

c

a

c

y

/

s

a

f

e

t

y

P

D

,

%

CpEfficacy PDSafety PD

Despite a relatively low peak-to-trough ratio, the drug does not have a therapeutic window.

100 mg dose

63

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Summary

PBPK modeling can be useful for multiple applications throughout the preclinical development process.Generating hypotheses and prioritizing experimentation Verifying that processes impacting absorption are

understood Selecting the most efficacious compound to move forward

The key is to use the model appropriately for the amount of validation that was possible.

PBPK modeling can be used to integrate multiple forms of preclinical data and to make a prediction of human PK/PD for both efficacy and safety end points.

64

P

h

a

r

m

a

c

e

u

t

i

c

a

l

s

Acknowledgements

Neil Parrott Thierry Lave Werner Rubas Ying Ou Dan Lu Pam Berry Paul Weller Karen Olocco Dimitrios Stefanidis Qingyan Hu Susan Larrabee

65

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sReferences Christensen H, Anders S, Holmboe AB, and Berg KJ. (2002).

Coadministration of grapefruit juice increases systemic exposure of diltiazem in healthy volunteers. Eur J Clin Pharmacol, 58: 515-520.

Connor, A, King, G, and Jones, K. Evaluation of Human Regional Bioavailability to Assess Whether Modified Release Development is Feasible AAPS 2008 Poster M1378

Hinderling, PH. (1997). Red blood cells: A neglected compartment in pharmacokinetics and pharmacodynamics. Pharmacological Reviews 49: 279-295.

Jamei M, Turner D, Yang J, Neuhoff S, Polak S., Rostami-Hodjegan A, and Tucker G. (In press 2009). Population-Based Mechanistic Prediction of Oral Drug Absorption. AAPS J.

Jones, HM, Parrott, N, Ohlenbusch, G, and Lav, T. (2006). Predicting pharmacokinetic food effects using biorelevant solubility media and physiologically based modelling. Clin. Pharmacokinet., 45, 1213-1226.

66

P

h

a

r

m

a

c

e

u

t

i

c

a

l

sReferences Martinez, MN, and Amidon, GL. (2002). A mechanistic approach to

understanding the factors affecting drug absorption: A review offundamentals. J Clin Pharmacol, 42:620-643.

Masimirembwa CM, Bredberg U, Andersson TB. (2003). Metabolic stability for drug discovery and development - Pharmacokinetic and biochemical challenges. Clin Pharmacokinet 42: 515-528.

Paine MF, Hart HL, Ludington SS, Haining RL, Rettie AE, and ZeldinDC. (2006). The human intestinal cytochrome P450 pie. DM&D34(5):880-886.

Peters, S. (2008). Evaluation of a generic physiologically basedpharmacokinetic model for lineshape analysis. Clin. Pharmacokinet., 47: 261-275.

Willmann, S, Schmitt, W, Keldenich, J, and Dressman, JB. (2003). A physiologic model for simulating gastrointestinal flow and drug absorption in rats. Pharm Res 20: 1766-1771.

Yang J, Jamei M, Rowland-Yeo K, Tucker G, and Rostami-HodjeganA. (2007). Prediction of Intestinal First-Pass Drug Metabolism. Current Drug Metabolism, 8: 676-684.